When you perform a linear regression, you can examine the R-square value, which is a goodness-of-fit statistic that indicates how well the response variable can be represented as a linear combination of the explanatory variables. But did you know that you can also go the other direction? Given a set
Do you know that you can create a vector that has a specific correlation with another vector? That is, given a vector, x, and a correlation coefficient, ρ, you can find a vector, y, such that corr(x, y) = ρ. The vectors x and y can have an arbitrary number
To help visualize regression models, SAS provides the EFFECTPLOT statement in several regression procedures and in PROC PLM, which is a general-purpose procedure for post-fitting analysis of linear models. When scoring and visualizing a model, it is important to use reasonable combinations of the explanatory variables for the visualization. When
Intuitively, the skewness of a unimodal distribution indicates whether a distribution is symmetric or not. If the right tail has more mass than the left tail, the distribution is "right skewed." If the left tail has more mass, the distribution is "left skewed." Thus, estimating skewness requires some estimates about
The expected value of a random variable is essentially a weighted mean over all possible values. You can compute it by summing (or integrating) a probability-weighted quantity over all possible values of the random variable. The expected value is a measure of the "center" of a probability distribution. You can
A fundamental principle of data analysis is that a statistic is an estimate of a parameter for the population. A statistic is calculated from a random sample. This leads to uncertainty in the estimate: a different random sample would have produced a different statistic. To quantify the uncertainty, SAS procedures
A previous article shows how to use a recursive formula to compute exact probabilities for the Poisson-binomial distribution. The recursive formula is an O(N2) computation, where N is the number of parameters for the Poisson-binomial (PB) distribution. If you have a distribution that has hundreds (or even thousands) of parameters,
When working with a probability distribution, it is useful to know how to compute four essential quantities: a random sample, the density function, the cumulative distribution function (CDF), and quantiles. I recently discussed the Poisson-binomial distribution and showed how to generate a random sample. This article shows how to compute
The Poisson-binomial distribution is a generalization of the binomial distribution. For the binomial distribution, you carry out N independent and identical Bernoulli trials. Each trial has a probability, p, of success. The total number of successes, which can be between 0 and N, is a binomial random variable. The distribution
Many textbooks and research papers present formulas that involve recurrence relations. Familiar examples include: The factorial function: Set Fact(0)=1 and define Fact(n) = n*Fact(n-1) for n > 0. The Fibonacci numbers: Set Fib(0)=1 and Fib(1)=1 and define Fib(n) = Fib(n-1) + Fib(n-2) for n > 1. The binomial coefficients (combinations
A previous article discussed how to solve regression problems in which the parameters are constrained to be a specified constant (such as B1 = 1) or are restricted to obey a linear equation such as B4 = –2*B2. In SAS, you can use the RESTRICT statement in PROC REG to
Matrix balancing is an interesting problem that has a long history. Matrix balancing refers to adjusting the cells of a frequency table to match known values of the row and column sums. One of the early algorithms for matrix balancing is known as the RAS algorithm, but it is also
On discussion forums, many SAS programmers ask about the best way to generate dummy variables for categorical variables. Well-meaning responders offer all sorts of advice, including writing your own DATA step program, sometimes mixed with macro programming. This article shows that the simplest and easiest way to generate dummy variables
Have you ever seen the "brain teaser" for children that shows a 4 x 4 grid and asks "how many squares of any size are in this grid?" To solve this problem, the reader must recognize that there are sixteen 1 x 1 squares, nine 2 x 2 squares, four 3 x 3 squares, and one 4 x 4 square.
When you write a program that simulates data from a statistical model, you should always check that the simulation code is correct. One way to do this is to generate a large simulated sample, estimate the parameters in the simulated data, and make sure that the estimates are close to
Last month a SAS programmer asked how to fit a multivariate Gaussian mixture model in SAS. For univariate data, you can use the FMM Procedure, which fits a large variety of finite mixture models. If your company is using SAS Viya, you can use the MBC or GMM procedures, which
I recently showed how to compute within-group multivariate statistics by using the SAS/IML language. However, a principal of good software design is to encapsulate functionality and write self-contained functions that compute and return the results. What is the best way to return multiple statistics from a SAS/IML module? A convenient
The multivariate normal distribution is used frequently in multivariate statistics and machine learning. In many applications, you need to evaluate the log-likelihood function in order to compare how well different models fit the data. The log-likelihood for a vector x is the natural logarithm of the multivariate normal (MVN) density
A previous article discusses the pooled variance for two or groups of univariate data. The pooled variance is often used during a t test of two independent samples. For multivariate data, the analogous concept is the pooled covariance matrix, which is an average of the sample covariance matrices of the
If you have ever run a Kolmogorov-Smirnov test for normality, you have encountered the Kolmogorov D statistic. The Kolmogorov D statistic is used to assess whether a random sample was drawn from a specified distribution. Although it is frequently used to test for normality, the statistic is "distribution free" in
If you have been learning about machine learning or mathematical statistics, you might have heard about the Kullback–Leibler divergence. The Kullback–Leibler divergence is a measure of dissimilarity between two probability distributions. It measures how much one distribution differs from a reference distribution. This article explains the Kullback–Leibler divergence and shows
A SAS/IML programmer asked about the best way to print multiple SAS/IML variables when each variable needs a different format. He wanted the output to resemble the "Parameter Estimates" table that is produced by PROC REG and other SAS/STAT procedures. This article shows four ways to print SAS/IML vectors in
Books about statistics and machine learning often discuss the tradeoff between bias and variance for an estimator. These discussions are often motivated by a sophisticated predictive model such as a regression or a decision tree. But the basic idea can be seen in much simpler situations. This article presents a
In a previous article, I discussed the binormal model for a binary classification problem. This model assumes a set of scores that are normally distributed for each population, and the mean of the scores for the Negative population is less than the mean of scores for the Positive population. I
Suppose that a data set contains a set of parameter values. For each row of parameters, you need to perform some computation. A recent discussion on the SAS Support Communities mentions an important point: if there are duplicate rows in the data, a program might repeat the same computation several
The ROC curve is a graphical method that summarizes how well a binary classifier can discriminate between two populations, often called the "negative" population (individuals who do not have a disease or characteristic) and the "positive" population (individuals who do have it). As shown in a previous article, there is
Are you a statistical programmer whose company has adopted SAS Viya? If so, you probably know that the DATA step can run in parallel in SAS Cloud Analytic Services (CAS). As Sekosky (2017) says, "running in a single thread in SAS is different from running in many threads in CAS."
The Johnson system (Johnson, 1949) contains a family of four distributions: the normal distribution, the lognormal distribution, the SB distribution (which models bounded distributions), and the SU distribution (which models unbounded distributions). Note that 'B' stands for 'bounded' and 'U' stands for 'unbounded.' A previous article explains the purpose of
From the early days of probability and statistics, researchers have tried to organize and categorize parametric probability distributions. For example, Pearson (1895, 1901, and 1916) developed a system of seven distributions, which was later called the Pearson system. The main idea behind a "system" of distributions is that for each
What is an efficient way to evaluate a multivariate quadratic polynomial in p variables? The answer is to use matrix computations! A multivariate quadratic polynomial can be written as the sum of a purely quadratic term (degree 2), a purely linear term (degree 1), and a constant term (degree 0).
